Influence of EEG channel reduction on lower limb motor imagery during electrical stimulation in healthy and paraplegic subjects

• Autor(a) principal: Júnior, Paulo Broniera
• Data de Publicação: 2022
• Outros Autores: Campos, Daniel Prado, Lazzaretti, André Eugênio, Nohama, Percy, Carvalho, Aparecido Augusto [UNESP], Krueger, Eddy, Teixeira, Marcelo Carvalho Minhoto [UNESP]
• Tipo de documento: Outros
• Idioma: eng
• Título da fonte: Repositório Institucional da UNESP
• Texto Completo: http://dx.doi.org/10.1007/s42600-021-00189-6; http://hdl.handle.net/11449/230319
• topic: EEG classification; Motor imagery; Signal classification

Purpose: Among a wide range of applications that make use of Brain-Computer Interfaces (BCIs), the pattern recognition of motor imagery (MI) to trigger neuromodulation systems in the control of functional movements has received increasing attention. In this work, we evaluate the effect of reducing the number of electroencephalography (EEG) channels in the performance of lower limbs’ motor imagery classification during the application of electrical stimulation (ES) in 20 Hz (ES20Hz), 35 Hz (ES35Hz), and 50 Hz (ES50Hz). Methods: Five subjects participated in the study, three healthy participants (average age of 28 years old) and two paraplegic volunteers, 43 and 47 years old, respectively. In total, each participant performed 90 repetitions of motor imagery of the lower limb with 11 EEG channels (10-10 configuration) under electrical stimulation. After the data acquisition, a systematic and artificial reduction in the number of EEG channels (decreasing from 11 to 1 and considering all cases 11 , 10 , … , 2 , 1) was applied to evaluate the offline classifiers. The pattern classification was performed using the following methods: (i) linear discriminant analysis (LDA), (ii) multilayer perceptron (MLP), and (iii) support vector machine (SVM). The accuracy performance of 11 different configurations regarding the EEG channels was obtained and studied. Results: The highest accuracy (86.5%) was obtained with the SVM classifier. There was no significant difference in the accuracy (median ± interquartile range) obtained with the 11-EEG channel configuration (SVM = ES20Hz: 78.50% ± 8.18%, ES35Hz: 77.80% ± 7.15%, ES50Hz: 75.80% ± 5.17% and LDA = ES20Hz: 69.40% ± 6.35%, ES35Hz: 69.00% ± 3.90%, ES50Hz: 67.30% ± 5.35%) and the 4-EEG channel configuration (SVM = ES20Hz: 72.10% ± 6.38%, ES35Hz: 72.20% ± 6.42%, ES50Hz: 68.20% ± 5.07% and LDA = ES20Hz: 63.30% ± 3.58%, ES35Hz: 63.60% ± 3.83%, ES50Hz: 60.46% ± 5.65%). Conclusion: These results are important for future neuroprosthesis implementations, as they indicate the possibility of simpler and more compact assistive technology control systems, being the main contribution of this work.